The Influence of Seismic Displacement Models on Spatial Prediction of Regional Earthquake-Induced Landslides


Regional Landslide Prediction is a Critical Step towards Achieving Community Resilience Against Landslides during Earthquakes. We Present a Predictive Methodology for Regional Earthquake-Induced Landslides that Uses a Pseudo-Three-Dimensional (Pseudo-3D) Procedure Implemented on a Digital Elevation Model (DEM). the Methodology Calculates a Seismic Displacement for Every Grid Cell in the DEM as the Basis for Predicting the Location and the Size of Landslide that is Determined using a Three-Dimensional (3D) Geometric Projection based on Topography. This "Pseudo-3D" Methodology for Predicting Landslides Outperforms Cell-Based Approaches. It Has the Advantage of Deriving Landslide Number and Size Characteristics of Individual Landslides. We Quantify the Influence of Seven Commonly Used Seismic Displacement Models: Four Rigid, Two Decoupled Flexible, and One Fully Coupled Flexible. Seismic Displacement Model Performance is Evaluated by Implementing a Prediction for the 2015 Mw6.5 Lefkada Earthquake Event in Greece for Which Precise Location and High-Resolution Landslide Volume Data Are Available for Model Validation and Ground Motion Intensity Measures Are Well Constrained by Nearby Ground Motion Recordings. the 3D Landslide Inventory Was Created by Topographically Differencing Pre-Existing DEMs from Post-Event Digital Surface Models Derived from Satellite and Unmanned Aerial Vehicle (UAV) Imagery. the Inventory Contains over 700 Landslides, Primarily Shallow Disrupted Slides and Rock Slides, Along with Occasional Occurrences of Deep-Seated, Rotational Failures. Qualitative Comparisons between Predicted and Mapped Landslides Show that All Seismic Displacement Models Predict Landslides that Broadly overlap Geospatially with Most Mapped Landslides and the Regions of High Mapped Landslide Densities Are Predicted. a Variety of Metrics Are Used to Quantify the Prediction Results: These Include Metrics Such as Landslide Number and Area Density, as Well as Correctly Predicted Ratio, overlapped Ratio, Centroid Distance, Ground Failure Capture, and Efficiency. the Results Show that the Rathje and Antonakos (Kmax, K-Velmax) Model and the Saygili and Rathje (PGA, PGV) Model Correctly Predict the Most Mapped Landslides and Obtain the Highest Prediction Efficiency, But at the Cost of the Highest overprediction. the Bray and Macedo Model Achieves the Best Performance in Predicting Landslide Number, Area Density, and Location Accuracy, But the Correct Prediction of the Mapped Landslides and the Prediction Efficiency Are Worse Than Other Seismic Displacement Models. the Results Highlight the Role of Seismic Displacement Models in the Prediction of Regional Earthquake-Induced Landslides and Guide Selection of the Seismic Displacement Model Depending on the Objective of the Prediction. Strength Parameters Also Affect Landslide Prediction, and We Demonstrate that All Evaluation Metrics Vary Widely Due to Variation of Cohesion and Friction Angle, Except for the Correctly Predicted Ratio. More Importantly, Variation in Prediction Results Due to the Selection of Seismic Displacement Models is Found to Be Comparable to Variation in Results Associated with Strength Parameter Uncertainty. a Critical Slope Parameter is Proposed and Defined as the Flattest Slope at Which Landslide Triggering May Occur for a Given Predictive Model. as Such, Seismic Displacement Models that Have Lower Critical Slopes Predict More Landslides and Differences in Critical Slopes for Each Seismic Displacement Model Provide a Parsimonious Explanation for the Different Performances of the Seismic Displacement Models in the Regional Earthquake-Induced Landslide Prediction.


Geosciences and Geological and Petroleum Engineering


National Aeronautics and Space Administration, Grant G17AP00088

Keywords and Phrases

Critical slope; Earthquake-induced landslides; Regional landslide prediction; Seismic displacement model; Uncertainty

International Standard Serial Number (ISSN)


Document Type

Article - Journal

Document Version


File Type





© 2023 Elsevier, All rights reserved.

Publication Date

01 Nov 2023